RE: [Xen-devel] [PATCH 1/2][VTD] pci mmconfig support to be used for ATS

["Kay, Allen M" <allen.m.kay@xxxxxxxxx>]

>
>  - Use the regular PCI config access functions, and if accesses above
>    the regular 255 registers are detected then use mmio (if
>    available) for accesses.  The regular PCI configuration space is
>    still always accessible using standard mechanisms.
>
>  - Introduce pci_ops integrated properly into Xen internals.  You may
>    even want to pci_ops on a per device or per bus basis.
>

I will look into to see which of the above method is cleaner to implement.

>I also see that you've modified struct pci_dev to accomomodate for
>specifc ATS needs.  The 'bus' and 'devfn' fields have been made non
>constant.  They were marked as 'const' for a reason --- modifying them
>leads to race conditions.  Further, you don't actually allocate the
>pci_dev using the proper allocate function.  If you don't want the ATS
>unit to be treated as a PCI device in Xen then don't use struct
>pci_dev.  Create a separate struct ats_unit containing the fields you
>need instead.
>

I will define a new ATS structure instead of using the existing pci_dev 
structure so that it will not disturb existing use model for pci_dev structure.

Allen
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel